[PATCH v2 0/2] nfsd: expose NFSv4 client state through Netlink

[Prabhakar Pujeri]

Administrators can inspect NFSv4 clients through the nfsd filesystem, but
collecting a server-wide view requires walking one directory and opening
multiple files for every client. That interface remains useful for detailed
inspection, but it is awkward for monitoring and automation.

Add a compact, privileged query to the existing nfsd Generic Netlink family:

  1. Dump basic NFSv4 client identity, address, lease, confirmation,
     reclaim, and callback state.
  2. Add O(1) snapshots of per-client session and stateid record counts.

The existing nfsd filesystem files and their output remain unchanged. A
companion nfs-utils v2 series adds `nfsdctl clients`; its cover Message-ID is:

  <cover.1787988918.git.prabhakar.pujeri@xxxxxxxx>

This reroll intentionally drops the grace-status patch. As discussed with
Chuck, that interface needs an explicitly scoped design that can represent
both current server-wide grace and possible per-export grace in the future.
I will address it in a separate proposal.

Changes since v1:

  - replace the raw sockaddr attribute with separate IPv4/IPv6 address,
    port, and optional IPv6 scope-ID attributes
  - add a nonzero client-table generation and
    genl_dump_check_consistent(), so churn marks the dump NLM_F_DUMP_INTR
  - replace session-list and stateid-IDR walks under cl_lock with maintained
    u64 counters, making every per-client snapshot O(1)
  - encode counts as variable-width Netlink uint attributes
  - drop the grace-status patch for a separately scoped design
  - rebase onto nfsd-testing at 65b583e874b3

Validation performed:

  - strict checkpatch on both patches: no errors, warnings, or checks
  - YNL regeneration and all 33 schema checks; generated files remained clean
  - headers_install and NFS admin-guide and Netlink-spec documentation builds
    with SPHINXOPTS=-W
  - full x86_64 GCC 16.1.1 kernel/modules build with W=1; no NFSD or
    changed-file warnings
  - focused NFSD W=1 build and Sparse over every fs/nfsd translation unit:
    no findings
  - two independent two-vCPU QEMU boots of the exact locally built kernel
  - each boot created one live NFSv4.2 client with one session, open stateid,
    lock stateid, and delegation stateid, and zero layout stateids
  - raw Netlink, nfsdctl, and /proc/fs/nfsd/clients state agreed in both runs
  - no kernel warning, Oops, panic, sanitizer report, lockup, or stall

Development assistance: an LLM assisted with review analysis, implementation,
commit-message drafting, and test-harness development; Sparse was used for
static analysis.

Assisted-by: LLM sparse

v1: https://lore.kernel.org/r/cover.1787638668.git.prabhakar.pujeri@xxxxxxxx

Prabhakar Pujeri (2):
  nfsd: add a Netlink dump of NFSv4 clients
  nfsd: report per-client NFSv4 state usage through Netlink

 .../admin-guide/nfs/nfsd-admin-interfaces.rst |  22 ++
 Documentation/netlink/specs/nfsd.yaml         | 118 ++++++
 fs/nfsd/netlink.c                             |   5 +
 fs/nfsd/netlink.h                             |   1 +
 fs/nfsd/netns.h                               |   1 +
 fs/nfsd/nfs4ctl.h                             |  10 +
 fs/nfsd/nfs4layouts.c                         |   2 +-
 fs/nfsd/nfs4state.c                           | 341 +++++++++++++++++-
 fs/nfsd/nfsctl.c                              |  13 +
 fs/nfsd/state.h                               |   7 +
 include/uapi/linux/nfsd_netlink.h             |  43 +++
 11 files changed, 553 insertions(+), 10 deletions(-)


base-commit: 65b583e874b371d539314272006833653364a06b
-- 
2.54.0